In recent years, the rapid development of large language models (LLMs) has brought new vitality into various domains, generating substantial social and economic benefits. However, this swift advancement has also introduced new security vulnerabilities. Jailbreaking, a form of attack that induces LLMs to produce harmful content through carefully crafted prompts, presents a significant challenge to the safe and trustworthy development of LLMs. Previous jailbreak methods primarily exploited the internal properties or capabilities of LLMs, such as optimization-based jailbreak approaches and methods that leveraged the model's context-learning abilities. In this paper, we introduce a novel jailbreak method, SQL Injection Jailbreak (SIJ), which targets the external properties of LLMs, specifically, the way LLMs construct input prompts. By injecting jailbreak information into user prompts, SIJ successfully induces the model to output harmful content. Our SIJ method achieves near 100\% attack success rates on five well-known open-source LLMs on the AdvBench, while incurring lower time costs compared to previous methods. More importantly, SIJ is the first method to exploit the external properties of LLMs for jailbreak attacks and exposes a new vulnerability in LLMs that urgently requires mitigation. To address this, we propose a simple defense method called Self-Reminder-Key to counter SIJ and demonstrate its effectiveness through experimental results. Our code is available at \href{https://github.com/weiyezhimeng/SQL-Injection-Jailbreak}{https://github.com/weiyezhimeng/SQL-Injection-Jailbreak}.
翻译:暂无翻译